CN-122018660-A - Heat abstractor of server

CN122018660ACN 122018660 ACN122018660 ACN 122018660ACN-122018660-A

Abstract

The application relates to a server heat dissipation device which comprises a heat exchange module, a state monitoring module and a control module, wherein the heat exchange module is arranged at a heat generating point of each processor and internally accommodates a heat exchange medium, the heat exchange medium dissipates heat of the processor through absorbing heat of the heat generating point, the state monitoring module comprises a processor state acquisition unit, the processor state acquisition unit is connected with a main board and is used for acquiring running state signals of each processor in real time, the control module is used for judging real-time heat dissipation requirements of the heat exchange module corresponding to each processor according to the acquired running state signals of the processor and outputting a flow regulation instruction, and the flow regulation module is used for receiving the flow regulation instruction and regulating flow of the heat exchange medium so as to regulate heat dissipation speed of each processor. The heat dissipation device can capture the start-stop state of the multiprocessor in real time, accurately adjust the flow of heat exchange medium of each heat exchange module, and realize heat dissipation according to requirements.

Inventors

LU GUI
WANG XIAOMENG
Xia Dailong

Assignees

华北电力大学

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. A server heat sink, the server including a motherboard and a plurality of processors, comprising: the heat exchange module is arranged at the heating point of each processor, and is internally provided with a heat exchange medium, and the heat exchange medium dissipates heat of the processors by absorbing the heat of the heating point; the state monitoring module comprises a processor state acquisition unit, wherein the processor state acquisition unit is connected with the main board and is used for acquiring running state signals of each processor in real time; The control module judges the real-time heat dissipation requirement of the heat exchange module corresponding to each processor according to the collected running state signals of the processors and outputs a flow regulation instruction; And the flow regulating module is used for receiving the flow regulating instruction and regulating the flow of the flowing heat exchange medium so as to regulate the heat dissipation speed of each processor.
2. The server heat sink of claim 1, wherein the heat exchange module comprises a cold plate disposed at a heat generating portion of each processor for dissipating heat from the processor; the state monitoring module comprises temperature sensors, wherein the temperature sensors are arranged on a liquid inlet branch and a liquid outlet branch of the cold plate and are used for collecting the liquid inlet temperature and the liquid outlet temperature of the heat exchange medium in real time; The control module is configured to preset a first temperature threshold T 1 and a second temperature threshold T 2 , calculate a liquid inlet and outlet temperature difference Δt _i of each cold plate according to the collected liquid inlet temperature and liquid outlet temperature, and calculate a target flow Q _obj_i of each cold plate according to the first temperature threshold T 1 , the second temperature threshold T 2 and the actual temperature difference Δt _i : When Δt _i >T 1 , the target flow Q _obj_i =Q 0 ×[1+(ΔT _i -T 1 )/T 1 ]; when Δt _i <T 2 , the target flow Q _obj_i =Q 0 ×[1-(T 2 -ΔT _i )/T 1 ]; When T 2 ≤ΔT_i≤T 1 , the target flow Q _obj_i =Q 0 ; Wherein Q 0 is the preset basic flow of the cold plate.
3. The server heat sink of claim 2, wherein, The flow regulating module comprises flow regulating valves which are arranged on liquid inlet branches of the cold plates and are used for regulating the opening degree according to the flow regulating instructions so as to regulate the flow of heat exchange media in the cold plates; the state monitoring module comprises flow sensors arranged on liquid inlet branches of all cold plates, and the flow sensors are used for collecting actual flow of heat exchange media; The control module is configured to collect the actual flow in the liquid inlet branch through the flow sensor, calculate the target opening theta _obj_i of the flow regulating valve of each liquid inlet branch according to the actual flow Q _act_i and the target flow Q _obj_i of the liquid inlet branch of each cold plate, send pulse signals to the corresponding flow regulating valve, and control the flow regulating valve to adjust the valve opening to the target value theta _obj_i .
4. The server heatsink of claim 3, wherein the control module is configured to preset a minimum maintenance flow rate Q min for a cold plate, the minimum maintenance flow rate Q min being less than a base flow rate Q 0 , and determine a target flow rate Q _obj_i =Q min for the cold plate when the collected processor operating state is inactive and T 2 ≤ΔT_i≤T 1 .
5. The server heat sink of claim 2 wherein the heat exchange module further comprises a main circulation pump, the output end of the main circulation pump is connected to each liquid inlet branch through a liquid inlet main path, the input end of the main circulation pump is connected to each liquid outlet branch through a liquid outlet main path, and the main circulation pump is used for driving heat exchange medium in the cold plate to perform circulation heat dissipation; The state monitoring module comprises a pressure sensor, wherein the pressure sensor is arranged on the liquid outlet main path and is used for collecting the actual pressure P _act of the heat exchange medium in the liquid outlet main path; The control module is configured to preset a maximum pressure threshold value P max and a minimum pressure threshold value P min , calculate a target rotating speed of the main circulating pump when P _act >P max or P _act <P min is carried out, and output a rotating speed adjusting instruction to the main circulating pump; The calculation formula of the target rotating speed is n _obj =n _act -delta P K, Δp=p _act -[(P min +P max )/2 ], where n _act is the actual rotational speed of the main circulation pump, n _objt is the target rotational speed of the main circulation pump, and K is a preset pressure-rotational speed adjustment coefficient.
6. The server heat sink according to any one of claims 1-5, further comprising an upper computer, wherein the upper computer is connected to the server; the control module is configured to: Presetting a safety temperature threshold T max and a safety duration threshold T max ; when judging that the collected liquid outlet temperature T _out_i of each cold plate is larger than a safety temperature threshold T max , sending overheat fault information of the processor to the upper computer; And sending a shutdown signal to the server when the liquid outlet temperature T _out_i of each cold plate is greater than the safety temperature threshold T max and the duration exceeds the safety duration threshold T max .
7. The server heat sink of claim 4, further comprising a spray heat module comprising a spray header disposed toward each secondary heat generating spot, the secondary heat generating spot being heat-dissipated by spraying a cooling medium to the secondary heat generating spot; the state monitoring module comprises a spraying temperature sensor, wherein the spraying temperature sensor is used for collecting the real-time temperature of each secondary heating point; The control module judges the real-time heat dissipation requirement of each secondary heating point according to the acquired real-time temperature of each secondary heating point and outputs the flow regulation instruction of each spray header; The flow regulating module comprises a spraying electromagnetic valve which is opened or disconnected according to flow regulating instructions of each spray header, so that the spray start and stop of each spray header are controlled to regulate the heat dissipation speed of each secondary heating point; the secondary heating points comprise other heating points except a processor in the server.
8. The server heatsink of claim 7, wherein the control module is configured to preset a spray start temperature threshold T _spray , a spray stop temperature threshold T _stop , and output a spray adjustment command according to the temperature of the secondary heat generating spot T _ j and the spray start temperature threshold T _spray , the spray stop temperature threshold T _stop : When T _j >T _spray is carried out, outputting an opening instruction to a spray electromagnetic valve corresponding to the secondary heating point; When T _j <T _stop is carried out, a closing instruction is output to the spraying electromagnetic valve corresponding to the secondary heating point; and when T _stop <T _j <T _spray is reached, judging that the spray adjusting instruction is not required to be output.
9. The server heat sink of claim 8, wherein the spray heat dissipation module further comprises a spray manifold, a spray main, a spray power unit, and a spray reservoir unit; The input end of the spraying main pipe is connected with the spraying liquid storage unit through the spraying power unit, after the output end of the spraying main pipe extends to the secondary heating point area, the spraying main pipe is branched to form spraying branch pipes corresponding to the secondary heating points, and the tail end of each spraying branch pipe is connected with the spraying heads of the secondary heating points; the spray solenoid valves are arranged on the spray branch pipes and are used for controlling the spray start and stop of the corresponding spray branch pipes; the state monitoring module comprises a spraying flow sensor and a spraying pressure sensor which are arranged on the spraying main pipe, and the spraying flow sensor and the spraying pressure sensor are respectively used for collecting the flow and the pressure of cooling medium in the spraying main pipe; The control module is configured to adjust the output power of the spray power unit according to the flow and pressure data of the cooling medium collected by the spray flow sensor and the spray pressure sensor, so as to adjust the flow and pressure of the cooling medium.
10. The server heat sink of claim 9, wherein the spray power unit comprises a spray pump, the flow of the cooling medium being adjusted by adjusting a rotational speed of the spray pump; the control module is configured to preset the target flow of each spray branch pipe and calculate the target flow Q _spray of the spray main pipe according to the opening quantity of the spray electromagnetic valve; The actual flow Q _spray_act of the spray main pipe and the target flow Q _spray are compared, namely the rotating speed of the spray pump is increased when Q _spray_act ＜Q _spray is carried out, and the rotating speed of the spray pump is reduced when Q _spray_act ＞Q _spray is carried out.

Description

Heat abstractor of server Technical Field The application belongs to the technical field of operation and maintenance of data centers, and relates to a server heat dissipation device. Background With the rapid development of artificial intelligence, big data and cloud computing technology, the computing power requirement of an AI server of a data center increases exponentially, and in order to meet the operation requirement of core scenes such as big model training, high-performance parallel computing and the like, the AI server usually adopts a multi-GPU cluster configuration scheme, and the number of GPUs carried by a single server can reach 8, 16 or even more. Because the GPU can generate extremely high heat flux when running at full load, and each GPU needs to keep synchronous and efficient operation, the heat dissipation effect directly determines the running stability, the calculation output efficiency and the hardware service life of the server. Therefore, efficient heat dissipation from a GPU operating at high speed is particularly important. The existing heat dissipation device mostly adopts a control mode of constant flow or simple uniform flow to conduct heat dissipation on each GPU, when the GPU is suddenly started and stopped, the heat dissipation requirement of the corresponding cold plate can be changed drastically instantaneously, full-scale flow heat dissipation is needed during operation, basic circulating flow only needs to be maintained during stopping, but the existing control cannot respond to the change in time, so that overheat risks of the running GPU occur due to insufficient flow, or energy waste is caused by flow redundancy of the running GPU. Therefore, a server heat dissipation device capable of precisely matching the dynamic heat dissipation requirement of the GPU is needed to solve the problems in the prior art. Disclosure of Invention The embodiment of the application provides a server heat dissipation device which can capture the start-stop state of a multiprocessor in real time, accurately adjust the heat exchange medium flow of each heat exchange module, realize heat dissipation according to needs and solve the problems that the existing heat dissipation device cannot respond to the overheat risk and the energy waste during shutdown caused by the dynamic start-stop of the processor in time. The embodiment provides a server heat dissipation device, which comprises a main board and a plurality of processors, wherein the heat dissipation device comprises a heat exchange module, a state monitoring module and a control module, wherein the heat exchange module is arranged at a heat generating point of each processor and internally accommodates a heat exchange medium, the heat exchange medium dissipates heat of the processors by absorbing heat of the heat generating point, the state monitoring module comprises a processor state acquisition unit, the processor state acquisition unit is connected with the main board and is used for acquiring running state signals of each processor in real time, the control module is used for judging real-time heat dissipation requirements of the heat exchange module corresponding to each processor according to the acquired running state signals of the processors and outputting a flow regulation instruction, and the flow regulation module is used for receiving the flow regulation instruction and regulating flow of the heat exchange medium so as to regulate heat dissipation speed of each processor. According to the technical scheme, the heat dissipation device can capture the start-stop state of the multiprocessor in real time, accurately adjust the heat exchange medium flow of each heat exchange module, achieve heat dissipation according to needs, avoid overheat and energy waste during shutdown of the processor, guarantee efficient and stable operation of the multiprocessor, and are high in adaptability and convenient to operate and maintain. In the embodiment, the heat exchange module comprises a cold plate, the cold plate is arranged at the heating position of each processor and used for radiating heat of the processors, the state monitoring module comprises temperature sensors, the temperature sensors are arranged on a liquid inlet branch and a liquid outlet branch of the cold plate and used for collecting the liquid inlet temperature and the liquid outlet temperature of a heat exchange medium in real time, the control module is configured to preset a first temperature threshold T 1 and a second temperature threshold T 2, calculate the liquid inlet and outlet temperature difference delta T _i of each cold plate according to the collected liquid inlet temperature and the collected liquid outlet temperature, and calculate the target flow Q _obj_i of each cold plate according to the first temperature threshold T 1, the second temperature threshold T 2 and the actual temperature difference delta T _i, wherein the target flow Q _obj_i=Q0×[1+(ΔT_i-T1)